WORKING
P A P E R
Longitudinal Perceptions
of Supportiveness from
Biological Fathers
and Maternal Mental
Health Problems
SARAH O. MEADOWS
WR-702
This product is part of the RAND
Labor and Population working
paper series. RAND working papers
are intended to share researchers’
latest findings and to solicit informal
peer review. They have been approved
for circulation by RAND Labor and
Population but have not been
formally edited or peer reviewed.
Unless otherwise indicated, working
papers can be quoted and cited
without permission of the author,
provided the source is clearly referred
to as a working paper. RAND’s
publications do not necessarily
reflect the opinions of its research
clients and sponsors.
is a registered trademark.
August 2009
This paper series made possible by the NIA funded RAND Center for the Study
of Aging (P30AG012815) and the NICHD funded RAND Population Research
Center (R24HD050906).
Longitudinal Perceptions of Supportiveness From Biological Fathers
and Maternal Mental Health Problems*
Sarah O. Meadows
Associate Social Scientist
RAND Corporation
1776 Main Street Santa
P.O. Box 2138
Santa Monica, CA 90407-2138
smeadows@rand.org
Phone: 310-393-0411, EXT 7916
Fax: 310-260-8160
*
The Fragile Families and Child Wellbeing Study is funded by the National Institute of Child Health and Human
Development (NICHD) and a consortium of private foundations. The author would like to thank Joan Tucker for
helpful comments on an earlier draft. This paper was presented at the 2008 Annual Population Association of
America Meeting in Detroit, Michigan. DO NOT CITE WITHOUT PERMISSION FROM AUTHOR.
Support Trajectories and Maternal Mental Health
2
Abstract
Together, increased popularity of the life course paradigm and availability of longitudinal data
addressing family relationships have raised important questions about how family processes
should be measured to ensure that longitudinal experiences are accurately captured. Using the
Fragile Families and Child Well-Being Study (N = 2,158) this paper focuses on trajectories of
mothers’ perceived supportiveness from a biological child’s father and mental health problems
five years after the birth. A mother’s relationship status with the father is significantly related to
her perceptions of supportiveness with married mothers reporting the highest levels of
supportiveness followed by mothers in cohabiting unions, romantic non-residential unions, and
finally, mothers not in a romantic relationship with the child’s father. Increasing slopes of
supportiveness are associated with fewer subsequent mental health problems, controlling for both
time-varying and time-invariant maternal and relationship characteristics. The discussion calls
attention to alternate ways in which longitudinal experiences can be modeled.
Key Words: social support, mental health, mothers, Fragile Families
Support Trajectories and Maternal Mental Health
3
Relationships are the fabric of social life. They give meaning to the world around us.
They help to buffer us from both catastrophic events and daily hassles while, at the same time,
frequently give rise to the very negative events that we experience. Certain aspects of social
relationships, especially the level of support we perceive from individuals who are closest to us,
appear to be more relevant to our well-being than others (House, Landis, & Umberson, 1988a).
Perceived support from parents and other close relatives, spouses, and friends can be a powerful
coping resource and protect our overall health and well-being. But support from close
relationships can also act as the source of unhappiness and dissatisfaction. Disappointment or
frustration with respect to perceptions of relationship supportiveness may lead to lower levels of
mental health. And this balance between helping and hurting can shift at any time.
Cornwell (2003) argues that, while the definition of social support is itself a static
concept, over time, it has dynamic properties. The key for researchers interested in social
relationships and health, then, is to find a concise and precise way to capture longitudinal
experiences of social support in a dynamic way that can reflect its growth, decay, or staticity
over time (House, Umberson, & Landis, 1988b). A number of options are available, each
having their own strengths and weaknesses. At a minimum, all require longitudinal data. The
study of family dynamics and functioning is one area of research that has seen recent growth in
the availability of such longitudinal data. In particular, the Fragile Families and Child WellBeing Study has allowed researchers to focus on relationship dynamics among non-traditional
family types. In an era when rates of both cohabitation and non-marital childbearing have seen
dramatic increases (Elwood & Jencks, 2004; Teachman, Tedrow, & Crowder, 2000), studying
these non-traditional families over time has also grown increasingly important. Given that the
Support Trajectories and Maternal Mental Health
4
majority of single-parent households are female-headed (Kreider, 2007), the bulk of this research
has focused on mothers.
Why, in particular, should social scientist and policy makers be concerned with maternal
perceptions of supportiveness? First, perceptions of support have shown to be very important for
mental health (Lin, Ye, & Ensel, 1999), even more so than actual received support (Wethington
& Kessler, 1986). Second, because mothers are primarily responsible for raising children and
spend more time with them than fathers (Sandberg & Hofferth, 2005), and because a mother’s
mental health affects her ability to parent (Lovejoy, Graczyk, OHare, & Neuman, 2000; Pachter,
Auinger, Palmer, & Weitzman, 2006), if we care about child well-being, we should care about
how mothers fare. Third, research shows that children raised in the context of low-quality, highconflict families fare worse than either those in high-quality, low-conflict families or in highconflict families that divorce (Amato, 2006; Hetherington & Kelly, 2003). This also holds true
for adults who remain in long-term, low-quality marriages (Hawkins & Booth, 2005)
This paper uses data from the Fragile Families and Child Wellbeing Study (FFCWS) to
explore long-term experiences of perceived support that mothers have with their child’s
biological father. These mothers are drawn primarily from an economically disadvantaged
population, where many families are living under separate roofs. Using latent growth curve
modeling, analyses will test whether trajectories of mothers’ perceived supportiveness are
associated with subsequent mental health problems. They will also examine whether certain
time-varying relationship characteristics are working through supportiveness trajectories to
influence mental health and whether certain time-invariant maternal characteristics, including a
history of mental health problems, mediate the association between supportiveness trajectories
and maternal mental health problems. Ultimately, the goal of the paper is to motivate a new way
Support Trajectories and Maternal Mental Health
5
of thinking about the measurement of long-term trends in variables and constructs that impact
well-being, one that pushes past static descriptions to focus on the dynamic aspect of social
relationships and the life course.
Background
The salubrious effect of social support, whether perceived or actual, has been well
documented (House, Landis, & Umberson, 1988a, b; Pearlin, Lieberman, Menaghan, & Mullan,
1981; Lin, Ye, & Ensel, 1999; Van der Zee, Bunk, & Sanderman, 1997; Wethington & Kessler,
1986). In general, this literature defines social support as the aspects of close social relationships
that provide an individual with a sense of belonging, respect, love, and security (Jacobsen, 1986).
The concept of supportiveness, then, is a direct measure of how much an individual perceives a
relationship to provide him or her with these feelings. This support may come from any number
of sources, with support from spouses receiving the most attention in the literature (Cutrona,
1996). Among married couples, spouses are typically the primary source of social support
(Beach et al., 1993; Dakof & Taylor, 1990) and support from other sources does not appear to
substitute for lack of spousal support (Coyne & DeLongis, 1986).
The role of spousal support in overall well-being may be especially salient for women,
given gender stereotypes which emphasize the importance of intimate relationships (Acitelli,
1996; Cutrona, 1996; Erickson, 1993).1 Acitelli and Antonucci (1994) find that perceptions of
social support in marriage are more predictive of relationship satisfaction and general well-being
among women than men. Although spousal support in particular is important, given recent
1
However, some existing literature has reported that men actually benefit more from marital relationships in terms
of support (Steil, 2000; but see Verhofstadt, Buysese, & Ickles, 2007 for contrasting evidence). Women are more
likely to give than to receive support and thus men are more often recipients (Belle, 1982). Yet, other studies have
reported a positive association between giving support and mental health (Väänvnen, Buunk, Kivimäki, Pentti, &
Vahtera, 2005). Ultimately, the literature is not completely in agreement about what these two findings mean for
gender differences in the protective effect of marriage.
Support Trajectories and Maternal Mental Health
6
changes in family formation and structure, as well as increases in non-marital births, support
from a child’s biological father, regardless of union status, has become a key determinant of
maternal mental health.
Father Support And Maternal Well-Being
Social support from biological fathers is especially pertinent to the sample used here,
namely mothers with young children. The majority are disadvantaged in terms of educational
attainment and financial resources. Many of these women do not live with and are not
romantically tied to the child’s biological father but rely on him for supplemental financial
assistance or co-parenting. For these families, lack of instrumental and emotional support can
have a profound effect on family functioning (Green & Rodgers, 2001; Henly, Danzinger, &
Offer, 2005; Meadows, forthcoming).
Motherhood, although a time of many positive and exciting changes, is also very
stressful. Maternal parenting stress has been linked to both increased rates of mental illness
among mothers (Crnic & Greenberg, 1990; Thompson, Merritt, Keith, Bennett, & Johndrow,
1993), as well as child behavior problems (Creasey & Jarvis, 1994; Cooper, McLanahan,
Meadows, & Brooks-Gunn, forthcoming; Crnic, Gaze, & Hoffman, 2005). According to family
stress theory, certain family level characteristics, such as social support from intimate partners,
can alleviate parenting stress. Existing research shows that mothers who are satisfied with the
level of supportiveness they perceive from partners report enhanced parenting, improved motherchild bonds, higher levels of self-esteem and self-efficacy (Baumrind, 1995; Belsky, 1993), and
perceive lower levels of parenting stress (Mulsow, Caldera, Pursely, Reifman, & Huston, 2002).
In sum, because of the increased pressures and decreased resources among disadvantaged, new
Support Trajectories and Maternal Mental Health
7
mothers, satisfaction with social support from fathers is an integral component of healthy
families.
Unfortunately, many of these families are not healthy, resulting in union dissolution
(Lichter, Qian, & Mellot, 2006). Clearly, something is changing within the family dynamic to
result in the end of the relationship. If a mother perceives a decline in support from the child’s
father she may become dissatisfied with the relationship and ultimately opt to leave it, resulting
in negative consequences for both mothers and children. Conversely, many mothers not in coresidential unions fully expect to move in with or marry their child’s father in the future (GibsonDavis, Edin, & McLanahan, 2005). Carlson, McLanahan, and England (2004) use the Fragile
Families and Child Wellbeing Study and find that perceptions of partner supportiveness are
powerful predictors of parents being in co-residential unions (i.e., marriage or cohabitation) one
year after a birth, regardless of relationship status at birth. From the mother’s standpoint, one
could then hypothesize that such union formations are the result of improving family dynamics
and levels of perceived supportiveness from fathers. So knowing something about how paternal
support changes over time may allow us to predict who is going to dissolve a union and who is
going to enter a union. But then the question is how to adequately measure social support
longitudinally.
Social Support as a Process
The life course paradigm asserts that individual biographies reflect long-term patterns of
change in social roles and statuses and that these trends can best be reflected through the use of
trajectories (Elder, 1985; Giele & Elder, 1998; George, 1999). Integral to this approach is the
dynamic aspect of social life. Life is a process, with individuals making their way through the
life course, navigating all of the important developmental periods from childhood to adolescence
Support Trajectories and Maternal Mental Health
8
to adulthood. Analogous to this aging process is the “life course” of social relationships, which
consists of changing perceptions and receipt of social support that parallel changes that occur in
relationships (House, Umberson, & Landis, 1988b).
Cornwell (2003) argues that social support is dynamic and that individuals can
experience support growth, support decay, and support staticity. Each trajectory has a different
implication for mental health. A support growth trajectory would suggest that individuals are
forming and strengthening intimate ties, which are protective of mental health. Entering a coresidential partnership or a marriage would be one such example (Kim & McHenry, 2002). In
contrast, support decay trajectory might imply that an individual has become increasingly
dissatisfied with the support he or she either perceives or receives from network ties. A divorce
or separation, for example, may result in declines in emotional or instrumental support from a
partner, in-laws, children, or friends (Kalmijn, 2007; Kalmijn & Broese van Groenou, 2005;
Terhell, Broese van Groenou, & Van Tilburg, 2004). A static support trajectory might imply that
an individual is satisfied with the support he or she perceives (and/or receives), or that social
conditions prevent that individual from making changes to social relationships that would alter
levels of social support.
Unfortunately, Cornwell (2003) falls one step short of actually measuring trajectories in a
truly dynamic fashion by relying on descriptive typologies of longitudinal trends in adolescent
perceptions of parental and peer support. Change in support is captured by simply subtracting
perceived support at time one from perceived support at time two, approximately one year later.
Typologies created this way are problematic for a number of reasons. First, they cannot account
for all of the conceptual components of trajectories—sequencing, timing, transitions, and
duration (Dupre & Meadows, 2007). Second, they cannot account for the rate of change in
Support Trajectories and Maternal Mental Health
9
social support. Third, they do not allow the researcher to incorporate other time-varying
characteristics of individuals and social relationships that may also affect well-being as the
trajectory is itself unfolding (i.e., marital status, employment status, household composition).
And fourth, collapsing information about change into group level variables results in a loss of
variance and may reduce explanatory power. Nonetheless, Cornwell finds that cumulative trends
in support are predictive of mental health. For his sample of adolescents, the deleterious effect
of support decay on depression was stronger than that of the salubrious effect of support growth.
Additional Predictors of Perceived Father Support
In the analyses that follow it will also be important to account for other factors that may
influence both a mother’s trajectories of perceived supportiveness from her child’s biological
father and her mental health status. Significant differences in many aspects of marital and
cohabiting relationships have been noted, especially in levels of commitment to the relationship
(Nock, 1995; Stanley, Whitton, & Markman, 2004) and mental health status (Marcussen, 2005).
Given that individuals in cohabiting couples may be less committed to the relationship than
individuals in marital unions it is possible that marital status is an important predictor of
supportiveness. Further, couples who are in romantic but non co-residential unions may be less
likely than married or cohabiting individuals to report satisfaction with the support received from
partners in those relationships given even lower levels of commitment.
A number of other maternal characteristics have been theoretically and empirically linked
to mental health problems and that may also affect maternal perceptions of supportiveness.
Individuals with higher socioeconomic status have a lower lifetime prevalence of depression than
lower status individuals (Blazer, Kessler, McGonagle, & Swartz, 1994) and the link between
poverty status and poor mental health has been well documented among adults (McLeod &
Support Trajectories and Maternal Mental Health
10
Nonnemaker, 1999). Ross and Wu (1995) have pointed to education, rather than income or race,
as the main mechanism through which SES affects health. Nonetheless, both education level and
race/ethnicity will be included in the models. Two additional variables tap into a mother’s stress
level, which may adversely affect both her mental health and ratings of relationship
supportiveness. These include the average number of hours the mother works and whether or not
she has had or adopted a new baby since the last interview. Employment characteristics,
especially, the number of hours a mother spends at work, have been linked to work-family
conflict, especially among mothers (Bellavia & Frone, 2004; Sulsky & Smith, 2005). And the
addition of another family member may increase the financial and time burdens that mothers
face.
Research Aims
To date, few studies have explicitly modeled the longitudinal association between social
support and adult well-being (see Olsen, Iversen & Sabroe, 1991; Turner, Hays, & Coates, 1993)
and none have explicitly modeled social support as a trajectory in a methodologically rigorous
way. Given the importance of support garnered from close, personal relationships for mental
health, especially that from a spouse, and the importance of maternal mental health for child
well-being, this study focuses on supportiveness trajectories mothers receive from biological
fathers. It addresses two specific research questions. First, how do mothers differ in their ratings
of perceived supportiveness based on relationship status with the biological father? It is
expected that supportiveness levels will decline prior to the end of the relationship and that
mothers no longer in a romantic relationship with the biological father will report lower levels of
perceived supportiveness. If this is the case, then static measures of perceived supportiveness
may not fully capture the dynamic nature of support over the course of a relationship.
Support Trajectories and Maternal Mental Health
11
Assessments early in the relationship may yield high levels of satisfaction whereas assessments
later on may reflect dissatisfaction leading up to dissolution. Thus, the second research question
asks: what is the association between longitudinal supportiveness experiences with the focal
child’s biological father and subsequent maternal mental health problems? To do this, the paper
will utilize latent growth curve models to examine the association between the intercept and
slope of perceived supportiveness trajectories and mental health problems five years after the
birth of a child. It is hypothesized that mothers who experience an increase in perceived
supportiveness trajectories will also report fewer mental health problems at year five than
mothers who experience an overall decline in perceived supportiveness.
In this stage of the analysis it will be important to include a number of maternal
characteristics to assess whether they may work through the trajectory itself to affect mental
health, in the case of time-varying variables, or whether they may mediate the association
between the trajectory parameters and maternal mental health, in the case of time-invariant
characteristics. This stage of the analysis will proceed in three steps, with each step adding a set
of characteristics. A baseline model examines the bivariate association between trajectories of
supportiveness and mental health problems. Additional models, building on each, successively
add: (1) time-varying characteristics of both the mother and her relationship with the biological
father that may affect her perceptions of his supportiveness (e.g., relationship status, poverty
status, hours worked, and additional births), (2) previous mental health problems, and (3) timeinvariant characteristics of the mother that may affect mental health (e.g., age, education,
race/ethnicity, and immigrant status).
Method
Data
Support Trajectories and Maternal Mental Health
12
Data come from the Fragile Families and Child Wellbeing Study (FFCWS), a national
longitudinal survey of parents and their children (Reichman, Teitler, Garfinkel, & McLanahan,
2001). The FFCWS consists of 4,898 children born in large U.S. cities, including 3,712 whose
parents were unmarried at birth. Maternal baseline interviews were conducted in-person, within
48-hours of the focal child’s birth, with follow-up interviews via telephone when the focal child
was one-, three-, and five-years of age. Response rates were 89 percent at year one, 86 percent at
year three, and 85 percent at year five.
Sample
Due to skip patterns in the data not all mothers responded to the supportiveness items.2 Mothers
who reported not being in a romantic relationship with the biological father and who had not
been in a romantic relationship with the father in previous were skipped out of the
supportiveness section of the survey. In many cases, these skip patterns are based on missing
information from mothers (i.e., those who move in and out of waves) because it is not clear if
they had ever been in a romantic relationship with the biological father. Supportiveness ratings
are available from 2,264 mothers across all four waves, with the bulk of the loss coming from
mothers who are missing data at year five (n=871, 17% of the total sample), followed by a group
of mothers who appear only at the baseline and year three interviews (n=496, 10% of the entire
sample) and a group of mothers who appear at baseline but never again (n=459, 9% of the entire
2
At baseline, supportiveness questions are not asked of 242 mothers. At year one, supportiveness questions are not
asked of mothers who are skipped out of the screener (n=344), who were currently not married or romantically
involved with the biological father and were not married to or romantically involved with the biological father at
baseline (n=388), and who were not interviewed (n=534). An additional 24 mothers refused to answer, or did not
know, one or more of the supportiveness items. At year three, supportiveness questions are not asked of mothers
who are skipped out of an initial screener (n=126), who were currently not married or involved with the biological
father and were not married to or in a romantic relationship with the biological father at baseline or year one (n=67),
and who were not interviewed (n=667). An additional 42 mothers refused to answer, or did not know, one or more
of the supportiveness items. At year five, supportiveness questions are not asked of mothers who are skipped out of
an initial screener (n=103), who were currently not married or romantically involved with the biological father and
were not married to or in a romantic relationship with the biological father at baseline, year one, or year three
(n=1,215), and who were not interviewed (n=759). An additional 21 mothers refused to answer, or did not know,
one or more of the supportiveness items.
Support Trajectories and Maternal Mental Health
13
sample). Data that is also missing, but not due to survey design, is discussed in the analysis
section.
Measures
Dependent Variable. A composite score for mental health problems is created by
summing three dichotomously coded items—heavy episodic drinking (i.e., binge drinking), illicit
drug use, and diagnosis of a major depressive episode—all of which are available at the year one,
year three, and year five interviews. Heavy episodic drinking is defined as consumption of at
least 5+ drinks in one sitting at least once in the previous month at the year one interview and 4+
drinks at the year three and five interviews. Roughly five percent of mothers at year one, 11
percent at year three, and 12 percent at year five report a recent episode of binge drinking (see
Table 1). Illicit drug use is defined as use of at least one illicit drug (sedatives, tranquilizers,
amphetamines, analgesics, inhalants, marijuana, cocaine, LSD/hallucinogens, or heroin) without
a prescription, in larger amounts than prescribed, or for longer than prescribed in the past month.
Two percent of mothers at year one and five percent at years three and five report recent illicit
drug use. Depression is measured using the Composite International Diagnostic Interview Short
Form (CIDI-SF) Version 1.0 November 1998 (see Kessler, Andrews, Mroczek, Ustum, &
Wittchen 1998). Scoring followed procedures outlined by the developers of the CIDI-SF to yield
12-month DSM-IV diagnoses of Major Depressive Episode (MDE) (American Psychiatric
Association, 1994; Walters, Kessler, Nelson, & Mroczek, 2002). Thirteen percent of mothers at
year one, 19 percent at year three, and 16 percent at year five meet the diagnostic criteria for
MDE. The mean mental health problem score across all mothers is .21 at year one, .35 at year
three, and .33 at year five.
Support Trajectories and Maternal Mental Health
14
Note that the CIDI depression measure is not obtained at the baseline interview because
of potential overlap with postpartum depressive symptoms. Similarly, comparable measures of
drug use and binge drinking are not available at the baseline interview. However, mothers are
asked to indicate whether they used drugs or alcohol during pregnancy. One percent of mothers
reported using drugs and two percent reported using alcohol during the prenatal period.
Independently, each of the three items in the composite mental health problems score has been
cited in existing studies as an indicator of poor mental health. Moreover, Aneshensel (2002) has
argued that disorder specific models provide a biased estimate of the impact of social factors and
stress on mental health when these factors may influence more than one health outcome. Indeed,
recent literature on the study of mental health has moved to including both internalizing (e.g.,
depression) and externalizing (e.g., alcohol use/abuse, violence) behaviors as indicators of
mental health problems (Umberson, Williams, & Anderson 2002). Combining depression, binge
drinking, and drug use into one measure of mental health problems maximizes the variability of
this construct within the sample and more fully captures the breadth of emotional distress than
may result from changes in family structure (see Meadows, McLanahan, & Brooks-Gunn, 2008;
Meadows, 2008).
Independent Variable. At the baseline, year one, year three, and year five interviews
mothers are given four statements related to the level of supportiveness they perceive from the
child’s biological father. The statements were: father is “fair and willing to compromise when
you have a disagreement” (fair), “expresses love and affection” (love), “criticizes you or your
ideas” (insult), and “encourages or helps you to do things that are important to you” (encourage).
Frequency was measured as “often” (3), “sometimes” (2), or “never” (1). The insult item was
reverse coded. Items are taken from the Multi-Dimensional Support Scale (Winefield,
Support Trajectories and Maternal Mental Health
15
Winefield, & Tiggemann, 2000). Descriptive statistics use a mean of valid items so long as
mothers responded to at least two of the supportiveness items. Cronbach alphas for the analysis
sample were .56 at baseline, .65 at year one, .72 at year three, and .77 at year five. Growth
models treat supportiveness as a latent construct, with categorical indicators (see Procedure
section). Using the mean from items that mothers responded to at each wave, the average
supportiveness ratings at baseline, year one, year three, and year five were 2.70, 2.58, 2.52, and
2.45, respectively (see Table 1).
[Insert Table 1 about here.]
The supportiveness items were asked of all mothers, regardless of their relationship status
with the father. When parents were no longer romantically involved, the statements asked
mothers to rate the supportiveness of the last month of the relationship. Because this protocol
may have resulted in mothers rating supportiveness of relationships that may have ended up to
five years ago (i.e., the period from birth to year five) an alternate specification of the
supportiveness measures recoded the observed supportiveness measures (i.e., fair, love, insult,
and encourage) of mothers who were not in a romantic relationship with the biological father as
zero for that wave.
Time-Varying Variables. The analyses include a number of time-varying variables.
Relationship status with the focal child’s biological father is measured at each wave: married
(reference category); cohabiting, but not married; in a romantic relationship, but not
cohabitating; and no relationship (not involved). Poverty is a dummy variable indicating whether
or not the mother’s income is at or below 100 percent of the federal poverty line at each wave.
Weekly hours worked tabulates the average number of hours the mother reported working during
the past year. Finally, a dummy variable indicates whether or not the mother had an additional
Support Trajectories and Maternal Mental Health
16
birth or adopted a new baby. With the exception of relationship quality and the indicator for a
new birth, which were assessed only at years one, three, and five, all other control variables are
available at each wave, including the baseline interview.
Time-Invariant Variables. Basic sociodemographic controls include mother’s age at
baseline (in years), education (less than high school, high school, some college, college and
above, with high school as the reference category), race/ethnicity (Black, White, Hispanic, Other
with White as the reference category), and immigrant status (dummy variable indicating mother
is a recent immigrant). Means and standard deviations of all control variables can be found in
Table 1.
Procedure
Because the paper is interested in the longitudinal association between maternal ratings of
supportiveness and mental health problems, a summary measure of supportiveness is needed.
Further, this measure should be able to handle latent constructs. Latent growth curve modeling
in a structural equation framework does both. This strategy assumes that mothers differ in initial
ratings of supportiveness based on certain time-invariant characteristics, and that variance in
subsequent growth (or decay) of supportiveness trajectories also varies by those characteristics.
The basic growth model is depicted in Figure 1.
[Insert Figure 1 here.]
Each mother’s trajectory is characterized by a unique intercept (α), linear, time-dependent
slope (β), and some measurement error (ε). Thus, the level one equation is as follows:
yit = αi + βit + εit
(Equation 1)
This equation represents within-individual (i) change over time (t). In order to incorporate the
time-varying variables into the model, Equation 1 must be modified as follows:
Support Trajectories and Maternal Mental Health
yit = αi + βit + γt wit + εit
17
(Equation 2)
The addition of the “γt wit” term represents the effect of each time (t) variable on supportiveness
at time (t) for each ith individual. In other words, each γ represents a perturbation from the latent
supportiveness trajectory associated with a particular marital status, poverty level, number of
weekly hours worked, or additional child at a specific point in time.
The second level of the growth model allows the random intercepts (αi) and slopes (βi) to
be a function of variables that differ across individuals (i) but do not change across time (t). This
level represents between-individual change over time. The level two equations are as follows:
αi = α0 + α1xi1 + α2xi2 + . . . αkxik + ui
(Equation 3)
βi = β0 + β1xi1 + β 2xi2 + . . . βkxik + vi
(Equation 4)
The intercept and slope for the supportiveness trajectory are directly regressed on these
characteristics to assess for potential group differences in the means of the growth factors.
What results from this latent growth model is a unique intercept, or starting point, and
trajectory, or course, of supportiveness for each mother in the sample. These trajectory
parameters, then, precisely summarize a mother’s longitudinal experience of her perceived
supportiveness from the focal child’s biological father. Intercepts and slopes can then be treated
as independent variables in a regression model where the dependent variable is mental health
problems at year five. The full model is depicted in Figure 1. In addition to the intercept and
slope, which take into account the full set of time-varying variables, the time-invariant maternal
characteristics (i.e., age, education, race/ethnicity, and immigrant status), as well as mental
health problems at year three, are also regressed on mental health problems at year five.
[Insert Figure 1 about here.]
Support Trajectories and Maternal Mental Health
18
All models are estimated using Mplus, Version 4.1 (Muthén & Muthén, 2006). Although
full information maximum likelihood (FIML) is available in Mplus, because mental health
problems is treated as a categorical variable the analytic sample is restricted to only mothers with
complete information on the x (i.e., exogenous) variables in the full model. This resulted in the
loss of an additional 106 mothers. Thus, the final analytic sample is 2,158. Possible attrition
bias is addressed by comparing the complete case sample (n=2,158) to the full sample (N=4,898)
on all of the variables in the final model to assess for substantive differences in these measures.
Compared to mothers in the analytic sample, those who are not in the analytic sample are
more disadvantaged (i.e., have lower levels of education and are more likely to live in poverty),
less likely to be a co-residential relationship (including both marriage and cohabitation) with the
biological father, more likely to be white and less likely to be black, and perceive higher levels
of perceived supportiveness at baseline, year one, and year three. However, all of the
aforementioned differences are substantively very small, always less than .10 units, and
differences grow smaller over time. Mental health differences are only apparent at year one (.22
versus .24 problems, p < 01) and at year three (.34 versus .37 problems, p<.05), where mothers
in the analytic sample report significantly fewer problems.
Given the skip patterns in the data it is not surprising that mothers who remain in the
complete case sample are more likely to co-reside with the biological father and to perceive
higher levels of supportiveness. Mothers with missing information at various wave typically
skipped the supportiveness items. If mental health problems are causing mothers to leave the
survey then any results using the complete case sample may be suspect due to reverse causation.
As such, it is important to compare the mental health of mothers who exit the survey in the year
prior to their departure to those who remain in all four waves. For example, mothers who exit
Support Trajectories and Maternal Mental Health
19
the survey after year one have a mental health problem score at year one but not years three or
five. Their year one score can be compared to the year one score of mothers who have valid
interviews at baseline and years one through five. Note that this is not a perfect test of selection
given the two-year duration between waves because mothers could experience a mental health
problem in the interim causing them to leave the survey. Nonetheless, it should give some
indication of whether mental health is related to attrition. No significant differences in selfreported mental health emerged between these groups of mothers. So it does not appear, based
on this test, that poor mental health is selecting mothers out of the survey.
Model fit is evaluated using the maximum likelihood ratio test statistic (χ2), which if
significant, indicates poor fit. Because models with sample sizes over 200 are frequently
significant, three supplemental measures of model fit are used: the root mean square error of
approximation (RMSEA), the Tucker Lewis Index (TLI), and the Comparative Fit Index (CFI).
Convention dictates that values of RMSEA below .05, and values of TLI and CFI close to 1.0
(Bollen & Curran, 2006), are indicative of good fit. All statistical tests referenced in the text are
two-tailed.
Results
Supportiveness Ratings. Recall that the first aim of the paper was to assess whether
mothers differ in their perceived ratings of supportiveness based on relationship status with the
focal child’s biological father. In doing so, it is possible to assess whether mothers “select”
themselves out of relationships based on lower perceived supportiveness. Table 2 presents these
results. What appears in the cells of the table are mean supportiveness ratings from mothers
based on their relationships statuses between waves. Mothers are classified as either in a
romantic relationship (i.e., married, cohabiting, or romantically dating) or not. For example, in
Support Trajectories and Maternal Mental Health
20
Row A, we see that mothers who were in romantic relationships with the child’s father at the
time of the birth rate the mean perceived supportiveness of the father as 2.72 whereas mothers
who were not in a romantic relationship with him at birth rate supportiveness significantly lower,
at 2.30 (p < .001). Remember that these mothers are using the last month of the relationship as
the basis for their ratings; however in this case, the last month of the relationship had to fall
somewhere between the conception of the child and the interview at birth (a rough maximum of
nine months ago).
[Insert Table 2 about here.]
Row B in Table 2 compares mothers who were in a romantic relationship with the father
at the baseline interview but ended that relationship sometime between the birth and the year one
interview. Although all of these women were coupled at baseline, there is already some
evidence of lower supportiveness perceptions among mothers who will end the relationship
within the next year (2.7 versus 2.59, p < .001). By the year one interview the difference has
grown larger, 2.63 versus 2.12 (p < .001), again suggesting that perceived relationship
supportiveness declined in the months leading up to the dissolution of the union.
Rows C and D show similar trends for mothers who end their romantic relationships with
biological fathers between years one and three and years three and five. In Row C, differences
between romantically involved and non-involved mothers are significant at every wave but are
much smaller at baseline (2.73 versus 2.63, p<.001). As the dissolution approaches, differences
grow larger. A similar pattern is also evident in Row D where differences are substantively
small at baseline but increase over time.
Of course a word of caution is required when interpreting these results because postdissolution mothers are retrospectively reporting supportiveness levels of relationships that were
Support Trajectories and Maternal Mental Health
21
not satisfactory. Thus it is not clear if low levels of satisfaction led to the end of the relationship
or whether the end of the relationship altered perceptions of supportiveness. Nonetheless, the fact
that a similar pattern exists across all mothers, regardless of when the relationship was dissolved,
suggests that dissatisfaction with support played an important role in the demise of these
relationships. Because differences in perceived supportiveness can be seen, in some cases, years
prior to the end of the relationship it is then important to think about how long-term relationship
experiences may influence well-being.
Supportiveness Trajectories. Before exploring a more complex model, it is useful to first
describe the basic growth model (i.e., no covariates) of supportiveness among all mothers.
Figure 2 depicts this basic growth model. Maternal supportiveness ratings at each interview are
treated as latent constructs and factor loadings and thresholds are constrained to be equal over
time to ensure measurement invariance (e.g., Fb = F1 = F3 = F5). Factor loadings for the
intercept are set to one at each wave, as is convention in growth models. Factors loadings for the
slope are set to one, three, and five at the year one, three, and five interviews, respectively, to
reflect time since the baseline interview. According to the model, the average mother’s
supportiveness trajectory starts at zero (ns) and decreases at a rate of .06 (p < .001) support units
per year. The results may appear nonsensical, but because significant variance exists around
both the intercept and slope mothers do have positive values of supportiveness. The correlation
between the intercept and slope indicates that mothers who start with higher levels of perceived
supportiveness will experience a slight increase in perceived supportiveness over time, although
this association is not significant (β = .001, ns).
[Insert Figure 2 about here.]
Support Trajectories and Maternal Mental Health
22
Longitudinal Supportiveness Experiences and Mental Health Problems. The second aim
of the paper was to assess the association between trends in maternal perceived supportiveness
and mental health problems at the year five interview. Results are presented in Table 3, in four
separate steps. Model 1 examines the basic association between supportiveness trajectory
parameters, that is the intercept and slope, and mental health problems at year five. Building on
Model 1, Model 2 adds controls for associations between mothers’ time-invariant characteristics
(TVVs) and their supportiveness trajectories (see Figure 1). In doing the model captures timespecific measurement error that is not captured by the growth factors themselves. Building on
the first two models, Model 3 adds a control for prior mental health problems at year three to see
if it mediates any of the association between the supportiveness trajectory parameters and mental
health at year five. And finally, Model 4 adds a direct association between mental health
problems at year five and the TIVs.
[Insert Table 3 about here.]
The first two columns of Table 3 (Model 1) report that a significant association exists
between supportiveness trajectories and mental health problems at year five. Specifically,
mothers who start their trajectories with higher levels of perceived supportiveness at the birth of
the child have fewer mental health problems five years later (β = -.55, p < .001). Further, an
increasing support trajectory (i.e., a positive slope) is associated with fewer mental health
problems at year five (β = -3.07, p < .001). These results suggest that long-term support
experiences do affect subsequent mental health.
Model 2, shown in columns three and four, adds the indirect association between the
TVVs and mental health problems at year five, via their associations with the latent
supportiveness constructs at baseline, year one, year three, and year five. Remember that the
Support Trajectories and Maternal Mental Health
23
TVVs include the mother’s marital status with respect to the biological father, her poverty status,
how many hours she works per week (on average), and whether she has given birth to or adopted
a new baby since the last interview. The associations between the supportiveness trajectory
parameters remain significant. Mothers who start out with higher levels of perceived
supportiveness experience fewer mental health problems at year five (β = -.61, p < .001) and
mothers who experience an increasing supportiveness trajectory also experience fewer mental
health problems at year five (β = -3.88, p < .001).
One of the TVVs appears to be a particularly salient for perceived supportiveness:
relationship status, especially having no involvement with the biological father. At all four
waves, mothers who report having no relationship with the biological father report much lower
levels of perceived supportiveness than mothers who are married to him. Table 2 showed that
relationship status is associated with significant differences in perceived supportiveness, with
mothers who leave relationships reporting lower levels of support in the month preceding the
dissolution. Given that this model specification simultaneously estimates the association
between relationship status and perceptions of supportiveness at each wave it cannot disentangle
the causal ordering.
Columns five and six show the results for Model 3, which adds a control for prior mental
health problems at year three, to assess whether a history of mental health problems can account
for the association between the supportiveness trajectory parameters and mental health at year
five. Not surprisingly, mothers who experience mental health problems at year three are likely to
experience problems at year five (β = .78, p<.001). The inclusion of previous problems reduces
the magnitude of the supportiveness intercept coefficient and renders it non-significant (β = .26). It does little, however, to change the slope coefficient (β = -3.10, p<.001). These results
Support Trajectories and Maternal Mental Health
24
suggest that in terms of mental health outcomes, where a mother starts her supportiveness
trajectory may not be as important as the path of that trajectory.
Finally, Model 4, shown in the last two columns of Table 3, incorporates the associations
between the TIVs and mental health problems at year five, including a mother’s age at baseline,
education level, her race/ethnicity, and her immigrant status. These variables are unable to
complete account for the association between the supportiveness trajectory parameters and
mental health problems at year five. In fact, the intercept parameter almost reaches statistical
significance (β = -.31, p<.10). The association between the slope and mental health problems at
year five weakens slightly, but still indicates that a decline in perceived supportiveness results in
more mental health problems at year five (β = -3.03, p<.001). It is not surprising that the TIVs
were unable to account for these associations given that only two, black and Hispanic, have
significant associations with mental health problems at year five.
Robustness Checks
Four alternate specifications of Model 4 were used to asses the robustness of the findings.
First, a model is estimated using the alternate coding of supportiveness indicators where mothers
no longer in romantic relationships with the biological father receive a score of zero on the
supportiveness indicators. Second, a subsample of the complete case sample containing only
mothers who were in romantic relationships with the biological father across all waves was
utilized. Mothers in this model were either married, cohabiting, or in romantic, non-residential
relationships. Third, mental health problems at year five were treated as a count variable. And
fourth, models were run using the three independent components of the overall mental health
problems measure: major depressive episodes, binge drinking, and illicit drug use.
Support Trajectories and Maternal Mental Health
25
Zero Supportiveness Ratings. Substantive results from models using the specification
where mothers not in romantic relationships with the biological father received a score of zero on
the observed measures of supportiveness (i.e., fair, love, insult, and encourage), did not differ
from Model 4 as reported in Table 3 (intercept β = -.69, p<.001 and slope β = -5.74, p<.001). It
should be noted that in some cases mothers in non-romantic relationships with the biological
father likely do perceive some amount of (positive) support from him so these results are
artificially decreasing supportiveness ratings overall, introducing zeros in the data, and as a
result, decreasing the variance.
Romantically Involved Mothers Only. A second version of Model 4 uses only mothers
who remained in romantic relationships with the biological father throughout the five-year
observation period. That is, these mothers were either married, cohabiting, or romantically
involved but non-resident. The total sample for this analysis is 1,395. Results form this model
support the findings presented in Table 4, Model 4. The intercept of the supportiveness
trajectory is not significant (β = -.05, ns) but the slope coefficient indicates that mothers who
experience a decline in perceived supportiveness from the biological father over time experience
more mental health problems at year five ( β = -5.99, p<.001). Clearly, the subgroup used in
this analysis is selective in that these are mothers who have been able to maintain a romantic
relationship with the father. One could assume that, on average, these mothers have a higher
levels of supportiveness than their peers who were unable to maintain such relationships.
However, it is revealing that changes in longitudinal support perceptions still had an association
with mental health among this select group of mothers.
Count Variable of Mental Health Problems. Supplementary analyses also estimated the
regression model for mental health problems using a Poisson distribution and a continuous
Support Trajectories and Maternal Mental Health
26
distribution. The substantive conclusions did not change. Because of the large proportion of
zeros among the mental health problem measure at year five the author also examined whether a
zero inflated Poisson (ZIP) regression model would be appropriate. The first step of this test was
to examine whether the mean of the outcome was (roughly) equal to its variance. Using the
analytic sample, the mental health problems measure at year five has a mean of .33 and a
variance of .36 (at year three µ = .34, σ2= .35 and at year one µ = .21, σ2= .19). The second step
of this test was to compare the distribution of each measure of the mental health outcome to a
known Poisson distribution using the mean of each outcome. Comparing these two figures
showed good fit between the expected distribution and the observed distribution for both the
FIML and restricted samples (results not shown). Taken together, these results suggest that there
is no problem with a preponderance of zeros and thus the ZIP model is not necessary.
Independent Mental Health Problems. Additional models analyzed each of the three
mental health composite items independently. For major depressive episodes, the results were
almost identical to those for overall mental health problems (intercept β = -.61, p<.001 and slope
β = -4.56, p<.001). Results were weaker for binge drinking, with the intercept of the
supportiveness trajectory not significantly associated with binge drinking at year five (β = .02,
ns) but the slope of the trajectory was significant and in the expected direction (β = -2.03, p<.05).
Finally, results for drug use are in line with the findings outlined above but coefficients do not
reach statistical significance (intercept β = -.17, ns and slope β = -1.28, ns). In light of the fact
that major depression is the most commonly reported mental health problem in the data and drug
use is the least common, these findings are to be expected. The experience of major depression
does appear to be driving the results, although substantively, binge drinking and illicit drug use
follow very similar patterns.
Support Trajectories and Maternal Mental Health
27
Discussion
This research sought to uncover whether long-term trends in maternal ratings of
supportiveness from her child’s biological father have an association with subsequent mental
health problems. It has argued that social relationships, as well as the support that is garnered
from them, can be viewed as processes and should be operationalized as such. Latent growth
models revealed that mothers who had higher levels of perceived supportiveness at the birth of
their child reported fewer mental health problems five years later. Similarly, mothers who
experienced growth in supportiveness over time also experienced fewer problems later.
However because perceived supportiveness can be influenced by a number of maternal
characteristics, statistical models also analyzed whether these factors could mediate the
associations between supportiveness trajectories and maternal mental health problems. The
negative association between supportiveness trajectory slopes and mental health problems was
robust to the inclusion of these variables into the model.
Results also showed that mothers who ended romantic relationships reported lower levels
of supportiveness during the last month of the union. Yet mothers who remained together the
longest before ending their relationships with fathers showed little difference in supportiveness
ratings from mothers who remained stably in romantic unions with biological fathers. The
differences, although statistically significant, were substantively small initially. However they
grew over the five year period in which women were observed. Taken together, these results
suggest that dissatisfaction at the end of a broken relationship may be the real driving force
behind the association between supportiveness trajectories and mental health problems observed
here. Despite the fact that union dissolution, a time-specific stressful event, may have an
immediate impact on mental health, long-term trends in social and emotional resources are still
Support Trajectories and Maternal Mental Health
28
likely to play an important role in individual well-being. Amassing support from family and
friends, even if it is only the perception of support, may well serve as a buffer when individuals
do encounter stress and strain.
Although it did little to affect the influence of supportiveness trajectory parameters on
subsequent mental health problems, a mother’s relationship status with the biological father did
affect how much support she received from the father. Not surprisingly, at all waves, married
mothers reported the highest levels of supportiveness followed by cohabiting mothers and
mothers in non-residential romantic relationships. By far, mothers who were no longer in any
type of relationship with the biological father reported the lowest levels of supportiveness at the
end of the relationship. Based on the models presented here it is not clear whether low levels of
support were a cause or a consequence of the end of the relationship. Given that these
excessively negative support ratings may have biased the results the analysis was re-run using
only mothers in romantic relationships, including marriages, cohabiting unions, and dating
relationships. These results, although using a much more selective sample yielded similar
results: declining support trajectories are associated with more mental health problems. Further,
an additional model used an alternate operationalization of support where mothers in nonromantic relationships automatically received a score of zero on the observed indicators (i.e.,
fair, love, insult, and encourage). Again these results yielded identical substantive results.
The results presented here may have important policy implications for families and child
well-being. Mothers who are no longer in romantic relationships with biological fathers report
lower levels of support. Social support, measured either as a snap-shot or longitudinally, is
associated with a mother’s mental health, which is then related to her ability to parent and
provide the resources needed to raise a healthy child. Thus the findings of this study suggest that
Support Trajectories and Maternal Mental Health
29
one way to improve both maternal and child mental health is to focus on relationship quality, and
perceived supportiveness, among non-co-resident parents (see Cowan, Cowan, Pruett, & Pruett,
2007). A recent study by Carlson and colleagues (2008) finds that positive co-parenting, as
perceived by mothers, is highly predictive of nonresident fathers’ involvement with children.
Obviously, in the optimal situation divorced or separated parents would “get along” for the sake
of the child because the impact of the parents’ relationship on child well-being works both
directly, through father involvement and co-parenting, and indirectly, through maternal mental
health. Thus one possible unintended consequence of the Healthy Marriage Initiative, designed
to foster relationship skills among romantically involved parents, is to cultivate continued
positive interactions between parents who end their unions (Dion, 2005).
Future Research
One of the primary goals of this research was to heighten awareness of the ways in which
social scientists can measure longitudinal processes in constructs important to well-being.
Indeed, the perception of supportiveness is a process because it waxes and wanes over time,
responding to normal changes in ever-evolving social relationships (Cornwell, 2003). In terms
of the salience of spousal support for mental health, existing research clearly suggests that it is
important (Waite, 1995), although men are believed to benefit more from spousal support and
women from friend support (Burleson, 2004; Cutrona, 1996). The idea that a partner provides
“built-in” emotional support is implicit in this literature. But how to measure that support,
especially over time, and how longitudinal trends in support are associated with well-being, is
less clear. For instance, is it more important, in terms of mental health, for perceptions of spousal
support to be constantly high (i.e., what is typically assume in the “built-in” theory of support) or
do perceptions need to grow over time to have an impact on well-being? Is a persistently low
Support Trajectories and Maternal Mental Health
30
level of spousal support better, worse, or the same as a decreasing trajectory of perceived
supportiveness? In a sample of adolescents, Cornwell (2003) found that the cumulative effect of
social support on depressive symptoms was significantly weaker if support trajectories were
categorized by staticity versus growth or decay. These findings could reflect a statistical artifact.
That is, with little or no variance in growth/decay, support may have little explanatory power.
Or the findings may reflect the fact that, in general, the status quo is good enough to maintain
current levels of mental health. If an individual is satisfied with his or her support, and has good
mental health, that level of well-being remains in tact. Unfortunately, this would also suggest
that if individuals have poor mental health and perceive little support there positive changes in
well-being will be equally unlikely.
More generally, future research should begin incorporate trajectories of other longitudinal
experiences, including stressful life experiences and daily strains, received expressive and
instrumental support, interaction with social networks, and lifestyle factors like diet and exercise.
Each of these constructs can be considered through a trajectory lens by increasing, decreasing, or
remaining constant over time. Having one set of parameters to express these trends may be a
useful way to summarize life experiences. And by simultaneously modeling multiple trajectories
researchers can test for mediating and cross-lagged associations, helping to disentangle timing
issues and move closer to a causal story.
Limitations
Although the Fragile Families Study is useful for studying patterns of well-being in
mothers across many relationship statuses, the sampling frame of mothers in large, urban areas
may limit its generalizability. Further, despite the use of a multifaceted question-set related to
relationship supportiveness, these questions are really only applicable to mothers who were
Support Trajectories and Maternal Mental Health
31
currently in a romantic relationship with the biological father at the specified wave. No current
measure of supportiveness, or relationship quality, is available for non-co-resident parents at
each wave. In some instances, mothers may be reporting on a relationship that ended four or five
years ago making it impossible to make definitive causal claims about the results. And even for
mothers who were in romantic relationships with the biological father it is possible that
perceptions of supportiveness changed multiple times in the time between interviews, a period
that could span up to two years. This highlights the importance of collecting more fine-grained
relationship data. In addition, no measures of perceived emotional support received from others,
including friends and family, is available. It is possible that declines in supportiveness from
fathers are offset by support garnered from others. Such counterbalancing trajectories are one
possible avenue for future research.
Conclusion
As social scientists, we need to think more about the best way to summarize long-term,
life-course processes, especially as more and more of our data sources become longitudinal.
This paper has focused on modeling longitudinal experiences mothers have with a child’s
biological father. Specifically, it tested whether trajectories of perceived supportiveness were
associated with subsequent mental health problems. This association does exist, but a mother’s
relationship status with the father is also an important determinant of the support she perceives.
Given the current fluidity of family formation in the United States today it is now especially
important to capture changing family and relationship dynamics, especially among parents with
young children. The results presented here call attention to alternate ways in which such
longitudinal experiences and processes can be modeled.
Support Trajectories and Maternal Mental Health
32
References
Acitelli, L.K. (1996). The neglected links between marital support and marital satisfaction. In
G. Pierce, B. Sarason, & I. Sarason (Eds.), Handbook of social support and the family
(pp. 83-103). New York: Plenum.
Acitelli, L.K, & Antonucci, T.K. (1994). Gender differences in the link between marital support
and satisfaction in older couples. Journal of Personality and Social Psychology, 67:688698.
Amato, P.R. (2006). Marital discord, divorce, and children’s well-being: Results from a 20-year
longitudinal study of two generations. In A. Clarke-Stewart & J. Dunn (Eds.), Families
count (pp. 179-202). New York: Cambridge University Press.
American Psychiatric Association. (1994). Diagnostic and Statistical Manual of Mental
Disorders, Fourth Edition. Washington, DC: American Psychiatric Association.
Aneshensel, C. (2002). Commentary: Answers and questions in the sociology of mental
health. Journal of Health and Social Behavior, 43, 236-46.
Baumrind, D. (1995). Child maltreatment and optimal care giving in social science (Vol.
1007).
New York: Garland.
Blazer, D. G, Kessler, R. C., McGonagle, K. A. & Swartz, M. S. (1994). The prevalence and
distribution of major depression in a national community sample: The National
Comorbidity Survey. American Journal of Psychiatry, 151, 979-986.
Beach, S. R. H., Martin, J. K., Blum, T. C., & Roman, P. M. (1993). Effects of marital and coworker relationships on negative affect: Testing the central role of marriage. American
Journal of Family Therapy, 21, 313–323.
Support Trajectories and Maternal Mental Health
33
Bellavia, G. & Frone, M. (2004). Work family conflict. In J. Barling, E.K. Kelloway, & M.R.
Frone (Eds.), Handbook of work stress (pp. 113-147). Thousand Oaks, CA: Sage.
Belle, D. (1982). The stress of caring: women as providers of social support. In L. Goldberger &
S. Breznitz (Eds.), Handbook of stress: Theoretical and clinical aspects (pp. 496–505).
New York: Free Press.
Belsky, J. (1993). The determinants of parenting: A process model. Child Development, 55, 8396.
Bollen, K.A., & and Curran, P.J. (2006). Latent Curve Models: A Structural Equation
Perspective. New Jersey: John Wiley and Sons, Inc.
Burleson, B. R. (2003). The experience and effects of emotional support: What the study of
cultural and gender differences can tell us about close relationships, emotion, and
interpersonal communication. Personal Relationships, 10, 1–23.
Carlson, M.J., McLanahan, S.S., & Brooks-Gunn, J. (2008). Coparenting and nonresident
fathers’ involvement with young children after a nonmarital birth. Demography, 45, 461488.
Carlson, M.J., McLanahan, S.S., & England, P. (2004). Union formation in fragile families.
Demography, 41, 237-261.
Cornwell, B. 2003. The dynamic properties of social support: decay, growth, and staticity,
and their effects on adolescent depression. Social Forces, 81, 953–978.
Cooper, C.E, McLanahan, S.S., Meadows, S.O., & Brooks-Gunn, J. Forthcoming. Family
structure transitions and maternal parenting stress. Journal of Marriage and Family.
Coyne, J. C., & DeLongis, A. (1986). Going beyond social support: The role of social
relationships in adaptation. Journal of Consulting and Clinical Psychology, 54, 454–460.
Support Trajectories and Maternal Mental Health
34
Cowan, C.P., Cowan, P.A., Pruett, M.K., & Pruett, K. (2007). An approach to preventing
coparenting conflict and divorce among low-income families: Strengthening couple
relationships and fostering fathers’ involvement. Family Processes, 46, 109-121.
Creasy, G.L., & Jarvis, P.A. (1994). Relationships between parenting stress and developmental
functioning among two-year-olds. Infant Behavior and Development, 17, 423-429.
Crnic, K. A., Gaze, C., & Hoffman, C. (2005). Cumulative parenting stress across the preschool
period: Relations to maternal parenting and child behaviour at age 5. Infant and Child
Development, 14, 117-132.
Crnic, K., & Greenberg, M. (1990). Minor parenting stresses with young children. Child
Development, 61, 1628-1637.
Cutrona, C.E. (1996). Social support as a determinant of relationship quality: The interplay
between negative and supportive behaviors. In G. Pierce, B. Sarason, & I. Sarason
(Eds.), Handbook of social support and family (pp. 173-194). New York: Plenum.
Dakof, G. A., & Taylor, S. E. (1990). Victims’ perception of social support: What is helpful for
whom? Journal of Personality and Social Psychology, 58(1), 80–89.
Dion, M.R. (2005). Healthy marriage programs: Learning what works.” The Future of
Children, 15, 137–256.
Dupre, M.E., & Meadows, S.O. (2007). Disaggregating the effects of marital trajectories on
health. Journal of Family Issues, 28, 623-652.
Elder, G. H., Jr. (1985). Perspectives on the life course. In G. H. Elder, Jr. (Ed.), Life course
dynamics: Trajectories and transitions, 1968-1980 (pp. 23-49). Ithaca, NY: Cornell
University Press.
Ellwood, D.T., & Jencks, C.J. (2004). The spread of single-parent families in the
Support Trajectories and Maternal Mental Health
35
United States since 1960. In D.P. Moynihan, T.M. Smeeding, & L. Rainwater (Eds.),
The future of the family (pp. 25-65). New York: Russell Sage.
Erickson, R.J. (1993). Reconceptualizing family work: The effect of emotion work on
perceptions of marital quality. Journal of Marriage and Family, 55, 888-900.
George, L. K. (1999). Life-course perspectives on mental health. In C. S. Aneshensel & J. C.
Phelan (Eds.), Handbook of the sociology of mental health (pp. 565-583). New York:
Kluwer Academic/Plenum.
Gibson-Davis, C., Edin, K., & McLanahan, S.S. (2005). High hopes but even higher
expectations: The retreat from marriage among low-income couples. Journal of Marriage
and Family, 67, 1301-1312.
Giele, J. Z., & Elder, G. H., Jr. (1998). Life course research: Development of a field. In J. Z.
Giele & G. H. Elder, Jr. (Eds.), Methods of life course research: Qualitative and
quantitative approaches (pp. 5-27). Thousand Oaks, CA: Sage.
Green, B.L., & Rodgers, A. (2001). Determinants of social support among low-income mothers:
A longitudinal analysis. American Journal of Community Psychology, 29, 419-441.
Hawkins, D.N., & Booth, A. (2005). Unhappily ever after: Effects of long-term, low-quality
marriages on well-being. Social Forces, 84, 451-471.
Henly, J.R., Danzinger, S.K., & Offer, S. (2005). The contribution of social support to the
material well-being of low-income families. Journal of Marriage and Family, 67, 122140.
Hetherington, E.M., & Kelly, J. (2003). For better or for worse: Divorce Reconsidered. New
York: W.W. Norton & Company.
House, J.S., Umberson, D., & Landis, K.R. (1988a). Social relationships and health. Science,
Support Trajectories and Maternal Mental Health
36
241, 540-545.
---------------------------------------------------. (1988b). Structures and processes of social support.
Annual Review of Sociology, 14, 293-318.
Jacobsen, T.E. (1986). Types and timing of social support. Journal of Health and Social
Behavior, 27, 250-264.
Kalmijn, M. (2007). Gender differences in the effects of divorce, widowhood, and remarriage
on intergenerational support: Does marriage protect fathers? Social Forces, 85, 10791104.
Kalmijn, M., & Broese van Groenou, M. I. (2005). Differential effects of divorce on social
integration. Journal of Social and Interpersonal Relationships, 22, 455-476.
Kessler R. C., Andrews G., Mroczek D., Ustun B., & Wittchen, H-U. (1998). The World Health
Organization Composite International Diagnostic Interview Short-Form (CIDI-SF).
International Journal of Methods in Psychiatric Research, 7, 171-185.
Kim, H.K., & McKenry, P.C. (2002). The relationship between marriage and psychological
well-being: A longitudinal analysis. Journal of Family Issues, 23, 885-911.
Kreider, R.M. (2007). Arrangements of children: 2004. Current Population Reports, P70-114.
U.S. Census Bureau, Washington, D.C.
Lichter, D.T., Qian, Z., & Mellott, L.M. (2006). Marriage or dissolution? Union transitions
among poor cohabiting women? Demography, 43, 223-240.
Lin, N., Ye, X., & Ensel, W.E. (1999). Social support and depressed mood. Journal of Health
and Social Behavior, 40, 344-359.
Lovejoy M.C., Graczyk, P.A., O’Hare, E., & Neuman, G. (2000). Maternal depression and
parenting behavior: a meta-analytic review. Clinical Psychology Review, 20, 561–592.
Support Trajectories and Maternal Mental Health
37
Marcussen, K. (2005). “Explaining Differences in Mental Health between Married and
Cohabiting Individuals.” Social Psychology Quarterly, 68, 239–57.
McLeod, J., & Nonnemaker, J. M. (1999). Social stratification and inequality. In C. S.
Aneshensel & J. C. Phelan (Eds.), Handbook of sociology of mental health (pp. 321-344).
New York: Kluwer.
Meadows, S.O. Forthcoming. Is It There When You Need It? Mismatch in Perception of
Future Availability and Subsequent Receipt of Instrumental Social Support. Journal of
Family Issues.
Meadows, S.O., McLanahan, S.S., & Brooks-Gunn, J. (2008). Family structure changes and
maternal health trajectories. American Sociological Review, 73, 314-334.
Mulsow, M., Caldera, Y.M., Pursley, M., Reifman, A., Huston, A.C. (2002). Multilevel factors
influencing maternal stress during the first three years. Journal of Marriage and Family,
64, 944-956.
Muthén, L.K., & Muthén, B.O. (1998-2006). Mplus user’s guide. Fourth Edition.
Los Angeles, CA: Muthén and Muthén.
Nock, S.L. (1995). “A Comparison of Marriages and Cohabiting Relationships.” Journal of
Family Issues,16, 53–76.
Olsen, O., Iversen, L. & Sabroe, S. (1991). Age and the operationalization of social support.
Social Science and Medicine, 32, 767-771.
Pacher, L.M., Auinger, P., Palmer, R., & Weitzman, M. (2006). Do parenting and the home
environment, maternal depression, neighborhood, and chronic poverty affect child
behavioral problems differently in different racial-ethnic groups? Pediatrics, 117, 13291338.
Support Trajectories and Maternal Mental Health
38
Pearlin, L.I., Menaghan, E.G., Lieberman, M.A., & Mullan, J.T. (1981). The stress process.
Journal of Health and Social Behavior, 22, 337-356.
Reichman, N. E., Teitler, J. O., Garfinkel, I., & McLanahan, S. S. (2001). Fragile Families:
Sample and design. Children and Youth Services, 23, 303-326.
Ross, C. E., & Wu, C. (1995). The links between education and health. American Sociological
Review, 60, 719-745.
Sandberg J.F., & Hofferth, S.L. (2005). Changes in children’s time with parents: A Correction.
Demography, 42, 391-395.
Stanely, S.M., Whitton, S.W., & Markman, H.J. 2004. Maybe I do: Interpersonal Commitment
and Premarital or Nonmarital Cohabitation. Journal of Family Issues, 25, 496-519.
Steil, J. M. (2000). Contemporary marriage: Still an unequal partnership. In C. Hendrick & S. S.
Hendrick (Eds.), Close relationships: A sourcebook (pp. 124–136). Thousand Oaks, CA:
Sage.
Sulsky, L., & Smith, C. (2005). Work stress. Belmont, CA: Thomas Wadsworth.
Teachman, J.D., Tedrow, L.M., & Crowder, K.D. (2000). The changing demography of
America’s families. Journal of Marriage and the Family, 6, 1234-46.
Terhell, E. L., Broese van Groenou, M. I., & Van Tilburg, T. (2004). Network dynamics in the
long-term period after divorce. Journal of Social and Personal Relationships, 21, 719–
738.
Thompson, R. J., Merritt, K. A., Keith, B. R., Bennett, L., & Johndrow, D. A. (1993). The role
of maternal stress and family functioning in maternal distress and mother-reported
psychological adjustment of non-referred children. Journal of Clinical Child Psychology, 22,
78-84.
Support Trajectories and Maternal Mental Health
39
Turner, H.A., Hays, R.B., & Coates, T.J. (1993). Determinants of social support among gay men:
the context of AIDS. Journal of Health and Social Behavior, 34, 37-53.
Umberson, D., Williams, K., & Anderson, K. (2002). Violent behavior: A measure
of emotional upset? Journal of Health and Social Behavior, 43, 189-206.
Väänänen, A., Buunk, B.P., Kivimäki, M., Pentti, J, & Vahtera, J. (2005). When is it better to
give than to receive: Long-term health effects of perceived reciprocity in support
exchange. Journal of Personality and Social Psychology, 89,176–193.
Van der Zee, K., Bunk, B., & Sanderman, R. (1997). Social support, locus of control, and
psychological well-being. Journal of Applied Social Psychology, 27, 1842–1859.
Verhofstadt, L.L., Buysse, A., & Ickles, W. (2007). Social support in couples: An examination
of gender differences using self-report and observational methods. Sex Roles, 57, 267282.
Waite, L.J. (1995). Does marriage matter? Demography, 32, 483-507.
Walters E.E., Kessler, R.C., Nelson, C.B., & Mroczek, D. (2002). Scoring the World Health
Organization’s Composite International Diagnostic Interview Short Form (CIDI-SF).
Retrieved February 13, 2007. (http://www3.who.int/cidi/CIDISFScoringMemo12-0302.pdf).
Wethington, E., & Kessler R.C. (1986). Perceived support, received support, and adjustment
to stressful life events. Journal of Health and Social Behavior, 27, 78-89.
Winefield, H.R., Winefield, A.H., & Tiggemann, M. (2000). The Multi- Dimensional Support
Scale. In K. Corcoran, & J. Fischer (Eds.), Measures for clinical practice: A
sourcebook. 3rd Ed. (2 vols.) (pp. 507-511). New York: Free Press.
Table 1. Means/Percentages and Standard Deviations for Analysis Variables, Complete Case Sample (N =
2,158)
Baseline
Mean/%
Mental Health
Binge Drinkinga
1.90%
Drug Usea
1.39%
Depression (CIDI)
na
Total MH Problems (0-3)
na
Relationship Characteristics
Supportivenessb (1-3)
2.70
2.54
Fairc (1-3)
d
Love (1-3)
2.82
Insulte (1-3)
2.70
f
Encourage (1-3)
2.76
Marital Status
Married
34.62%
Cohabiting
42.22%
Romantic
21.32%
Not Involved
1.85%
New Baby
na
Demographics
Race
Black
42.86%
White
26.04%
Hispanic
26.55%
Other
4.26%
g
16.45%
Immigrant
Age
25.83
Education
Less HS
27.85%
HS
30.40%
Some College
26.69%
College
15.07%
h
29.47%
Poverty
Hours Worked
33.07
Year-One
SD
Mean/%
na
5.52%
1.67%
13.44%
.21
Year-Three
SD
Mean/%
(.44)
10.90%
5.26%
18.55%
.35
(.32)
(.57)
(.41)
(.51)
(.47)
2.8
2.38
2.76
2.54
2.67
(.41)
(.69)
(.48)
(.63)
(.57)
2.52
2.34
2.67
2.49
2.58
na
42.96%
41.33%
6.95%
8.76%
4.73%
46.39%
29.66%
3.71%
20.25%
25.21%
34.11%
33.70
32.76%
(13.89) 35.14
Year-Five
SD
Mean/%
SD
(.59)
12.05%
5.33%
15.94%
.33
(.59)
(.46)
(.64)
(.57)
(.64)
(.63)
2.45
2.29
2.57
2.45
2.49
(.51)
(.69)
(.63)
(.67)
(.67)
46.99%
21.08%
2.78%
29.15%
32.385%
(6.0)
(13.11)
(12.80)
31.58%
35.79
(13.05)
Notes: Variable range in parenthesis, if applicable. SD = Standard Deviation
a
At baseline, both refer to use during pregnancy. b Mean maternal rating of relationship supportiveness from biological father. If
mother and father are no longer romantically involved score reflects supportiveness from the last month of the relationship.
Higher scores indicate higher perceived supportiveness. c Father is fair and willing to comprise during disagreements. Higher
score indicates higher frequency. d Father expresses affection or love. Higher score indicates higher frequency. e Father insults or
criticizes ideas. Reverse coded. Higher score indicates lower frequency. f Father encourages or helps do things important to
mother. Higher score indicates higher frequency. g Dummy variable indicating mother is of immigrant status. h Dummy variable
indicating mother is at or below 100% of poverty level.
Table 2. Comparison of Pre- and Post-Romantic Involvement Maternal Ratings of Relationship
Supportiveness with the Biological Father, Complete Case Sample (n=2,158).
Relationship Exit
Between:
A. Conception and Baseline
Involved (n=2,115)
Not Involved (n=43)
B. Baseline and Year One
Involved (n=1,969)
Not Involved (n=189)
C. Year On and Year Three
Involved (n=1,721)
Not Involved (n=437)
D. Year Three and Year Five
Involved (n=1,529)
Not Involved (n=629)
Baseline
Year One
Year Three
Year Five
Mean
Mean
Mean
Mean
2.72***
2.30
2.72***
2.59
2.63***
2.12
2.73***
2.63
2.63***
2.42
2.62***
2.10
2.73***
2.65
2.64***
2.47
2.62****
2.28
2.60***
2.07
Notes: Romantic relationships are defined as married, cohabiting, or romantically dating. Shaded cells indicate
mothers’ supportiveness ratings of the last month couples were together.
†p < .10 *p < .05 **p <.01 ***p < .001
Support Trajectories and Maternal Mental Health
42
Table 3. Latent Growth Curve Model of Mothers’ Perceived Supportiveness Trajectories on
Mental Health Problems at Year Five (N=2,158)
Model 1
β
Model 2
SD
β
Model 3
SD
β
Model 4
SD
β
SD
-.31†
-3.03***
.75***
.17
.78
.04
Year Five Mental Health Problems
Supportiveness
Intercept
-.55***
Slope
-3.07***
MHP Year Three
TIVs
Age at Baseline
Education
Less High School
Some College
College Plus
Rae/Ethnicity
Black
Hispanic
Other
Immigrant Status
TVVs
Cohabiting
Romantic
Not Involved
Poverty
Hours Worked
TVVs
Cohabiting
Romantic
Not Involved
Poverty
Hours Worked
New Child
TVVs
Cohabiting
Romantic
Not Involved
Poverty
Hours Worked
New Child
TVVs
Cohabiting
Romantic
Not Involved
.14
.49
-.61***
-3.88***
.16
.73
-.26
-3.10***
.78***
.16
.78
.04
Baseline Supportiveness
.07*
.03
.07*
.03
.02
.03
.01
.03
-.42***
.05
-.42*** .05
-.01
.02
.001
.001
.001†
.001
-.01
.02
Year One Supportiveness
-.05
.04
-.05
.04
-.09†
.05
-.09*
.05
-.59***
.04
-.59*** .04
-.06*
.02
-.06**
.02
-.001†
.001
-.001†
.001
.02
.04
.01
.04
Year Three Supportiveness
-.15***
.04
-.15*** .04
-.14*
.07
-.15*
.07
-.63***
.04
-.62*** .04
.001
.03
.002
.03
-.001
.001
-.001
.001
.003
.03
.01
.03
Year Five Supportiveness
-.21***
.05
-.19*** .04
-.34***
.08
-.33*** .08
-.78***
.04
-.77*** .04
.01
.01
.08
.08
-.08
.08
.08
.12
-.33***
-.31***
-.14
-.19†
.08
.09
.18
.10
.06*
.002
-.43***
-.01
.001†
.03
.03
.05
.02
001
-.05
-.09†
-.58***
-.06*
-.001
-.002
.03
.05
.04
.03
.001
.04
-.16***
-.17*
-.64***
.004
-.001
.002
.04
.07
.04
.03
.001
.03
-.20*** .04
-.34*** .08
-.79*** .04
Support Trajectories and Maternal Mental Health
Poverty
Hours Worked
New Child
Model Fit
χ2 (df)a
RMSEA
CFI/TLI
-.001
.001
.04
251.09*** (48)
.044
. 894/.958
.03
.001
.03
439.77*** (173)
.027
.929/.943
.01
.001
.03
.03
.001
.03
.002
.001
.02
43
.03
.001
.03
611.72*** (184) 679.44*** (227)
.891/.912
.891/.906
.033
.030
Notes: Models are nested. TVVs = Time-Varying Variables. TIVs = Time-Invariant Variables. Model 1 includes
the supportiveness growth curve parameters (intercept and slope). Model 2 adds the association between the TVVs
and the latent support constructs of supportiveness at baseline, year one, year three, and year five. Model 3 adds the
association between mental health problems at year three and year five. Model 4 adds the association between the
TIVs and mental health at year five (i.e., age at baseline, education level, race/ethnicity, and immigrant status).
a
Chi-square values should not be used to compare fit across models with categorical endogenous variables.
†p < .10 *p < .05 **p <.01 ***p < .001
Support Trajectories and Maternal Mental Health
44
Figure 1. Full Growth Model of Maternal Ratings of Relationship Supportiveness with the
Biological Father and Mental Health Problems at Year Five.
TVVs
FB
LB
IB
TVVs
EB
L1
F1
Support
Baseline
I1
TVVs
E1
F3
Support
Year One
I3
E3
F5
Support
Year Three
Support
Intercept
TIVs
L3
TVVs
L5
I5
Support
Year Five
Support
Slope
Mental
Health
Problems
Year Five
Mental
Health
Problems
Year Three
Notes: TVVs = Time-Varying Variables (marital status, relationship quality, poverty, weekly hours worked,
additional child); TIVs = Time-Invariant Variables (mother’s age at baseline, educational status, race/ethnicity,
immigrant status); F = Fair; L = Love; I = Insult; E = Encourage. Subscripts indicate data wave (B = Baseline, 1 =
Year One; 3 = Year Three; 5 = Year Five. Model 1 includes the supportiveness growth curve parameters (intercept
and slope). Model 2 adds the association between the TVVs and the latent support constructs of supportiveness at
baseline, year one, year three, and year five. Model 3 adds the association between mental health problems at year
three and year five. Model 4 adds the association between the TIVs and mental health at year five. (i.e., age at
baseline, education level, race/ethnicity, and immigrant status).
E5
Support Trajectories and Maternal Mental Health
45
Figure 2. Base Latent Growth Model of Maternal Ratings of Supportiveness with the Biological
Father.
FB
LB
IB
.780 .818
EB
.751
F1
L1
E1
L3
F3
Support
Year One
0
I3
E3
F5
L5
I5
1.00
Support
Baseline
1
I1
1
Support
Year Three
1
1
Support
Intercept
α = .000
3
Support
Year Five
5
Support
Slope
β = -.06***
.001
Notes: F = Fair; L = Love; I = Insult; E = Encourage. Subscripts indicate data wave (B = Baseline, 1 = Year One; 3
= Year Three; 5 = Year Five). Factor loadings are constrained to be equal across waves. Error variances and
correlations among observed supportiveness indicators are not shown. Variances around both the intercept and
slope are significant.
Model fit: χ2 = 394.99, df = 97, p < .001; RMSEA = .038; CFI/TLI = .966/.958; n = 2,158.
†p < .10 *p < .05 **p <.01 ***p < .001
E5